An Interference Recognition Embedded People Counting Method Exploiting Occlusion Information

Radar-based people counting methods can extract crowd information from radar echo data without being affected by privacy or lighting issues. However, target occlusion and interference in the scene pose challenges to radar-based people counting. To address this problem, we propose an interference recognition embedded people counting approach exploiting occlusion information. First, we generate target point clouds and perform occlusion pattern recognition. Then, we use occlusion prior information to assist target tracking and obtain accurate target trajectories. Finally, we extract temporal features from the trajectories and apply a long short-term memory network to identify interference targets, achieving an average recognition accuracy of 97.39%. Meanwhile, we count people according to the category and location information of the target trajectories. The experimental results show that the proposed method can effectively solve the issues of target occlusion and interference, enhancing the robustness of people counting. Specially, the accuracy of daily people counting for passageway entry and exit reaches 96%.